【作者】 张昀；

【导师】 沈海斌；

【作者基本信息】 浙江大学 ， 电路与系统， 2008， 硕士

【摘要】 唇读(lip—reading/speech—reading),可以通过观察说话者的口型变化,“读出”或“部分读出”其所说的内容。唇读研究的目的是利用视觉信道信息补充听觉信道信息,提高计算机系统的理解力。唇读技术源于听力弱者或者听力障碍者学习、了解正常人的表达的一种技巧,它亦可用于特定场合的信息获取(如情报等)。如今,该技术被广泛应用于语音识别、身份识别、人机智能接口以及多媒体系统等领域。唇部检测作为唇读系统的首要环节,主要包含两个方面的内容,第一是在说话者环境中检测出脸部的人脸识别技术,第二是在已经识别出的人脸图像基础上的唇部识别技术。由于人脸识别技术已经有比较成熟的检测方法,本文主要研究在人脸彩色图像基础上的唇部检测算法。本文基于人脸的彩色图像,对不同人种的唇色和肤色的R,G,B分布进行了细致研究,提出一种基于唇色肤色色度差异的唇部检测算法。该算法充分利用了R,G,B三个分量的分布关系定位唇部,简单高效,具有较好的鲁棒性,适用于白色人种和黄色人种。本文将该算法与经典的Chromatic Feature Extraction算法和Red Exclusion算法进行比较,实验表明,本文算法在诸多方面有较大的进步。最后本文将所提出的算法用硬件描述语言加以实现,结果表明,新算法在速度,硬件开销上都符合嵌入式系统的应用要求。更多还原

【Abstract】 Lip-reading or speech-reading systems can understand or partially understand what a speaker says via his/her lip movements. The lip-reading research aims at compensating for the audio channel information by video information channel in order to enhance the computers’ intelligent level. The technology of lip-reading that comes from the skills which are often used by poor listeners to understand what others say can also be made full use of to get information in certain cases (e.g. intelligence). Nowadays, this technology is widely used in the realm of voice identification, identity identification, human-computer intelligent interface and multi-media etc.Lip detecting acted as one of the most important steps of lip-reading systems contains 2 facets, one is human face identification technology using which to detect the face of a speaker, the other is lip-detecting technology on the base of a human face that is already detected by face identification technology. In this paper, research was mainly carried out on the base of color images of human faces because the methodology of detecting face is mature enough in current times.In this paper, the R, G, B distribution of lip and skin color was detailedly researched in different race based on human face chromatic images and a new lip detecting algorithm was brought forward. This simple, effective, robust new algorithm that was fit for both white and yellow race made full use of the diversity of R, G, B distribution diversity between lip and skin to locate the lip area. The comparisons were done among this new algorithm and other two famous algorithm called Chromatic Feature Extraction algorithm and Red Exclusion algorithm. The results of experiments showed this new algorithm had many improved features.Finally, hardware description language was used to implement the new algorithm which was proved to be fit for embedded systems by the guide line of processing speed and hardware spending.更多还原